Saturday, April 14, 2018

"Bike" or "Bicycle" or Weak Automated Recognition

If you search for "bike" in the Library of Congress materials in Flickr, there is a considerable amount of noise. It seems to be a result of Flickr's automated recognition of certain things and automated adding of tags that is often incorrect when it identifies "bikes" in particular.

Rimouski -- handling coffins of victims  (LOC)

The tags for the above photo added by Flickr include "bike" and "vehicle."

Rose & Lena Krautly and Jitney bus  (LOC)

Again, "bike" and "vehicle" (and also "outdoors)" - it appears it is the spokes that confuse things.

Abernathy kids  (LOC)

A motorcycle is arguably a kind of bike - tags added by Flickr here were again "vehicle" and "bike."

Crippled German soldier [on bicycle] (LOC)

This is in fact a bicycle - this photo has same automated tags "vehicle" and "bike."

If however you search for "bicycle" in the Library of Congress metadata, a different, but complementary, pattern emerges.

Fogler  (LOC)

As with many photographs in Flickr of bicycles, the spokes are not that visible. Here there is no automated added tag for either "vehicle" or "bike." Instead there is a tag that was manually added for "bicycle" providing access. And in fact, the overlap between items with both "bike" and "bicycle" is quite small, only 14 items and for several of those, both the "bike" and "bicycle" tags were manually added.

In conclusion - searching for historic photographs of bikes and bicycles can be a noisy process.

No comments:

Post a Comment